Search Results

1 result for “rmsprop”

Formulae (1)

E[g^2]_t = \gamma E[g^2]_{t-1} + (1-\gamma)g_t^2

Scales the learning rate by a running average of recent squared gradients.